question about flow shift
#7
by zzzero - opened
"Why did the optimal 'flow shift' need to be set much larger when using the Flux2 VAE in your previous experiments compared to the Flux VAE（https://bfl.ai/research/representation-comparison）, and why is the 'flow shift' here still set to the default value, the same as the Flux VAE?"
Problably because this is a new model trained from scratch with a better architecture, but thats just my geuss.